Mohamed Abdel-Aty,Kirolos Haleem,Patrick Kerr,Joseph Santos,Xuesong Wang
Abstract:Crash prediction models are commonly used in safety analysis for relating crashes to the geometric, roadway and traffic characteristics, as well as for prediction purposes. However, safety researchers rarely consider more applied techniques for predicting crashes, rather than statistical models. In this study, a recently-developed Web-based application is introduced that could be helpful for practitioners. This application provides crash profile tables for various intersection configurations and different traffic conditions at signalized and unsignalized intersections. Crash patterns by intersection category was identified based on the 3 most recent available years of crash data, which include the expected number of crashes, standard deviation and highest percentiles by type, severity, lighting condition, surface condition, month, week, and hourly distributions within a certain county, FDOT district and the overall state. These profiles could serve as a crash patterns’ manual that could be used as reference values, hence assisting in identifying intersections with specific problems, e.g., high number of fatal crashes or high number of rear-end crashes, etc. Extensive design, traffic and crash data were collected from close to 1500 signalized and 2500 unsignalized intersections to develop these profiles. The whole system was then converted into Web application that is both automatically updatable and dynamic. Users can add more intersections to the application and upload the 3 most recent available crash data easily. It is envisioned that eventually all intersections in Florida will be included as part of the application